Über die Struktur der mittleren Entropie

From MaRDI portal





As in the paper of \textit{K. R. Parthasarathy} [Ill. J. Math. 5, 299--305 (1961; Zbl 0100.33903)], but independently, the problem solved is that of the integral representation of the entropy and transmission (information) rates in the case of finite alphabet stationary information sources and channels. Also the method used in both papers is based on the theory of \textit{N. Krylov} and \textit{N. Bogolyubov} [Ann. Math. (2) 38, 65--113 (1937; Zbl 0016.08604)]. Let \( A= \{1, \ldots, a\} \) be an alphabet, let \(\Omega = \left\{\omega=\left(\cdots, \omega_{-1}, \omega_{0}, \omega_{1}, \ldots\right): \omega_{t} \in A\right\}\), and let \(T\) be the shift transformation of \(\Omega\) onto \( \Omega\): \((T \omega)_{t}=\omega_{t+1} \). Let \[ _s(\omega)_t = \{\eta: \eta \in \Omega, \eta_s = \omega_s, \ldots, \eta_t = \omega_t\} \quad (s \leq t \text{ integers}). \] For given \(s, t\) there are \(a^{t-s+1}\) different \({}_s(\omega)_t\) constituting the set \(Z(s, t)\) of atoms of a finite \(\sigma\)-algebra \( B(s, t)\). Let \(B\) the \(\sigma\)-algebra generated by \(\cup_{s \leq t} B(s, t)\). In the sequel we are concerned with the measurable space \( (\Omega, B) \). Let \( V_0\) be the class of all stationary probability measures on \((\Omega, B)\). If \(p \in V_0\), then the entropy rate is defined by \(H(p) = \lim _t t^{-1} H(1, t ; p) \), where \(\displaystyle H(s, t ; p) = -\sum_{E \in Z(s, t)} p(E) \log p(E) \) is the entropy corresponding to \(p\) and \(B(s, t)\). According to the McMillan's limit theorem, the function \[ h_t(\omega ; p) \{=-t^{-1} \log p ({}_1(\omega)_t) \text{ for }p({}_1(\omega)_t)>0, \, =0 \text{ otherwise } \} \] converges in \( L^1(p) \) to a function \(h(\omega ; p) \) defined almost everywhere \( [p] \). The author proves (Theorem 1) that there exists such a set \(E \in B\) that \( p(E)=1 \) for every \(p \in V_0\) and that to every point \(\omega \in E\) there corresponds an ergodic \(m_{\omega} \in V_0\) such that \(m_{\omega}=m_{T \omega} \) and such that the invariant (with respect to \(T\) ) function \(h(\omega)\) \(\{= H(m_{\omega})\) if \( (\omega \in E) \), \( =0 \) otherwise \( \} \) is \( B_0\)-measurable, where \(B_0\) is the \( \sigma \)-algebra of the invariant sets of \(B\). Further, the following assertions are valid: 1. \( H(p)=\int h d p \) for every \(p \in V_0\); 2. \(\lim_t h_t(\omega ; p)=h(\omega)\) in \( L^1(p) \) for every \(p \in V_0\). A similar assertion as assertion 1 is proved for the transmission rate (Theorem 2 ).











This page was built for publication: Über die Struktur der mittleren Entropie

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q775393)